BUG: Categorical[dt64tz].to_numpy() losing tz (#38136)

* BUG: Categorical[dt64tz].to_numpy() losing tz

* Update pandas/core/arrays/categorical.py

Co-authored-by: gfyoung <[email protected]>

* Avoid FutureWarning

* whatsnew

Co-authored-by: gfyoung <[email protected]>

Author: jbrockmendel

doc/source/whatsnew/v1.2.0.rst

@@ -533,7 +533,7 @@ Categorical
 - Bug in :meth:`Categorical.__setitem__` that incorrectly raised when trying to set a tuple value (:issue:`20439`)
 - Bug in :meth:`CategoricalIndex.equals` incorrectly casting non-category entries to ``np.nan`` (:issue:`37667`)
 - Bug in :meth:`CategoricalIndex.where` incorrectly setting non-category entries to ``np.nan`` instead of raising ``TypeError`` (:issue:`37977`)
--
+- Bug in :meth:`Categorical.to_numpy` and ``np.array(categorical)`` with timezone-aware ``datetime64`` categories incorrectly dropping the timezone information instead of casting to object dtype (:issue:`38136`)
 
 Datetimelike
 ^^^^^^^^^^^^

pandas/conftest.py

@@ -320,6 +320,16 @@ def index_or_series(request):
 index_or_series2 = index_or_series
 
 
+@pytest.fixture(
+    params=[pd.Index, pd.Series, pd.array], ids=["index", "series", "array"]
+)
+def index_or_series_or_array(request):
+    """
+    Fixture to parametrize over Index, Series, and ExtensionArray
+    """
+    return request.param
+
+
 @pytest.fixture
 def dict_subclass():
     """

pandas/core/arrays/categorical.py

@@ -1269,15 +1269,13 @@ def __array__(self, dtype=None) -> np.ndarray:
             if dtype==None (default), the same dtype as
             categorical.categories.dtype.
         """
-        ret = take_1d(self.categories.values, self._codes)
+        ret = take_1d(self.categories._values, self._codes)
         if dtype and not is_dtype_equal(dtype, self.categories.dtype):
             return np.asarray(ret, dtype)
-        if is_extension_array_dtype(ret):
-            # When we're a Categorical[ExtensionArray], like Interval,
-            # we need to ensure __array__ get's all the way to an
-            # ndarray.
-            ret = np.asarray(ret)
-        return ret
+        # When we're a Categorical[ExtensionArray], like Interval,
+        # we need to ensure __array__ gets all the way to an
+        # ndarray.
+        return np.asarray(ret)
 
     def __array_ufunc__(self, ufunc, method, *inputs, **kwargs):
         # for binary ops, use our custom dunder methods

pandas/core/series.py

@@ -84,7 +84,13 @@
 from pandas.core.generic import NDFrame
 from pandas.core.indexers import deprecate_ndim_indexing, unpack_1tuple
 from pandas.core.indexes.accessors import CombinedDatetimelikeProperties
-from pandas.core.indexes.api import Float64Index, Index, MultiIndex, ensure_index
+from pandas.core.indexes.api import (
+    CategoricalIndex,
+    Float64Index,
+    Index,
+    MultiIndex,
+    ensure_index,
+)
 import pandas.core.indexes.base as ibase
 from pandas.core.indexes.datetimes import DatetimeIndex
 from pandas.core.indexes.period import PeriodIndex
@@ -412,7 +418,13 @@ def _set_axis(self, axis: int, labels, fastpath: bool = False) -> None:
             labels = ensure_index(labels)
 
         if labels._is_all_dates:
-            if not isinstance(labels, (DatetimeIndex, PeriodIndex, TimedeltaIndex)):
+            deep_labels = labels
+            if isinstance(labels, CategoricalIndex):
+                deep_labels = labels.categories
+
+            if not isinstance(
+                deep_labels, (DatetimeIndex, PeriodIndex, TimedeltaIndex)
+            ):
                 try:
                     labels = DatetimeIndex(labels)
                     # need to set here because we changed the index

pandas/tests/base/test_conversion.py

@@ -316,18 +316,34 @@ def test_array_multiindex_raises():
             TimedeltaArray(np.array([0, 3600000000000], dtype="i8"), freq="H"),
             np.array([0, 3600000000000], dtype="m8[ns]"),
         ),
+        # GH#26406 tz is preserved in Categorical[dt64tz]
+        (
+            pd.Categorical(pd.date_range("2016-01-01", periods=2, tz="US/Pacific")),
+            np.array(
+                [
+                    Timestamp("2016-01-01", tz="US/Pacific"),
+                    Timestamp("2016-01-02", tz="US/Pacific"),
+                ]
+            ),
+        ),
     ],
 )
-def test_to_numpy(array, expected, index_or_series):
-    box = index_or_series
+def test_to_numpy(array, expected, index_or_series_or_array):
+    box = index_or_series_or_array
     thing = box(array)
 
     if array.dtype.name in ("Int64", "Sparse[int64, 0]") and box is pd.Index:
         pytest.skip(f"No index type for {array.dtype}")
 
+    if array.dtype.name == "int64" and box is pd.array:
+        pytest.xfail("thing is Int64 and to_numpy() returns object")
+
     result = thing.to_numpy()
     tm.assert_numpy_array_equal(result, expected)
 
+    result = np.asarray(thing)
+    tm.assert_numpy_array_equal(result, expected)
+
 
 @pytest.mark.parametrize("as_series", [True, False])
 @pytest.mark.parametrize(